fix(detector): detect aliased bmm self output

[prasannakotyal]

Summary

• add SELF_MATMUL_OUTPUT for aliased torch.bmm used on an unsqueezed input and its transpose
• add sentinel replay coverage for the companion visible red batch
• register the self-matmul pattern as a hard auto-filter rule

Target

KernelGuard-Red-Submission: 793

Validation

• UV_CACHE_DIR=/tmp/uvcache uv run python -m py_compile kernelguard.py
• red 793 sample: classification=hacked, should_filter=true, pattern SELF_MATMUL_OUTPUT
• red 792/794/795/796 samples: classification=hacked, should_filter=true
• UV_CACHE_DIR=/tmp/uvcache uv run python ../../kernelguard_bypasses/eval_blue_patch.py kernelguard.py clean fixtures remain should_filter=False

[github-actions]

KernelGuard Blue Evaluation

• Evaluation job: 203
• Blue submission: 175
• Result: passed
• Target red submission caught: True
• Validation suite: TP 20/20, FP 20/20
• Surgicalness score: 1.0
• Workflow run: https://github.com/SinatrasC/kernelguard/actions/runs/25506140172

[SinatrasC]

Thanks for the KernelGuard Flywheel Campaign contribution. This PR is now superseded by the consolidated rule-family implementation in #273, which folds this detector coverage together with the related passing-eval variants.